text
stringlengths
0
8.13M
III. ACASESTUDY:RESOURCEALLOCATIONIN quantumcomputeroperatorcanrequestthedeploymentofnew
DISTRIBUTEDQUANTUMCOMPUTING quantum computers, or the utilization of on-demand quantum
computingmayberequired.Shortly,quantumcomputingfrom
A. Allocating Resources in Distributed Quantum Computing
multipleorganizationscanbesharedinwhichoneorganization
In the proposed distributed quantum computing framework, can borrow or buy quantum computing time from another
a quantum computer operator serves quantum computation organization temporarily. Such a scenario is inspired by the
tasks by distributing each of them to one or multiple quantum presentconventionalcloudcomputingparadigm.Forexample,
computers that can work collaboratively to accomplish the Amazon Braket offers on-demand quantum cloud computing
quantum computation. Quantum tasks may require the num- services [13]. For this option, the quantum computer operator
ber of qubits, which each quantum computer provides. The installs and configures the on-demand quantum computers to
works collaboratively with the deployed quantum computers. first part is the first-stage cost of using the reserved quantum
Inaddition,thecostofusingthedeployedquantumcomputers computers and the second part is the expected second-stage
islessthaninstallingthenewquantumcomputer.Utilizingthe cost that consists of the computing power unit and Bell pair
deployedquantumcomputerwillcostthecomputingpowerof costs of the deployed quantum computers including the on-
the quantum computer and the Bell pair of the two connected demand quantum computer deployment under the set of the
quantum computers. However, using an on-demand quantum possible realized demand of quantum tasks, computing power
computer is more expensive. The objective of the quantum of quantum computers, and fidelity of the entangled qubits.
computer operator is to satisfy the quantum computing tasks The constraints of the stochastic programming model are
while minimizing the total deployment cost. the selection of the utilized quantum computers, both the
deployed and the on-demand quantum computers being able
C. Uncertainty of Distributed Quantum Computing
to complete the computational tasks, and the utilization of the
Whiledeterminingthebestresourceallocationindistributed computingpowerunderthelinkcapacity,withconsideringthe
quantum computing, uncertainties occur, which can be classi- uncertainty.
fied into three types as follows:
E. Experimental Results
• First, the actual demands are unknown when the option
of deploying the quantum computer is made. Different 1) Parameter Setting: We consider the system model of
applicationssuchasminimization,dimensionalreduction, quantum computing where the quantum computer operator
and machine learning problems may request various consists of 10 deployed quantum computers. We set the cost
qubits [14]. For example, up to 30 qubits are used in values, which is measured in normalized monetary, for using
machine learning problems [14]. the deployed quantum computer, qubits, and Bell pairs to be
• Second, the precise availability of the quantum computer 5000, 1000, and 450, respectively. All quantum computers
and its qubits is uncertain since they might be reserved have 257 qubit capacities and identical costs. The cost and
for other purposes or because the quantum computer’s computing power of new quantum computer deployment are
backend may not support all of them [15]. For example, 25000 and 127, respectively. Costs for both on-demand and
only5of10quantumcomputersareavailabletocompute quantumcomputersaredeterminedbasedon[1],[13].Tosolve
quantum tasks. theproposedstochasticmodel,weconsidertwoscenarios.The
• Third, the fidelity of the entangled qubits is also not first scenario is that the demand of the quantum task is 10,
known exactly due to the degradation of the entangled the available computing power of the quantum computers is
qubits in distributed quantum computing [12]. For exam- 127qubits,andthefidelityoftheentangledqubitsinquantum
ple, if the fidelity is 0.5, it means that 50% efficiency of networksis1(i.e.,thebestperformance).Thesecondscenario
the entangled qubits can be achieved. is that there is no demand, no availability of qubits, and zero
Therefore,anadaptiveresourceallocationapproachisrequired fidelity of the entangled qubits in quantum networks (i.e., the
to efficiently provision quantum computers to tackle quantum worst performance). We assume the default probability values
computational tasks with dynamic sizes under uncertainty with 0.8 and 0.2, respectively.
of the computing power of quantum computers and fidelity 2) Impact of Probability of Scenarios: We vary the prob-
fidelity of the entangled qubits. ability of the first scenario, which corresponds to having the
demand, the availability of the quantum task, and the best
D. The Proposed Approach
performance for the entangled qubits. The cost breakdown is
Inthedeterministicresourceallocationforcomputingquan- shown in Fig. 3b. We note that when the probability of the
tum tasks in distributed quantum computing, the required scenario is equal to or less than 0.2, the new quantum com-
demand of quantum tasks, the computational power of quan- puter should be deployed. The deployed quantum computer is
tum computers, and the fidelity of the entangled qubits are utilized over the new quantum computer when the probability
exactly known by the quantum computer operator. Therefore, of the scenario is higher as it has the demand of the quantum
quantum computers can be certainly deployed and the on- task,computingpowerofquantumcomputers,andthefidelity.
demandquantumcomputerdeploymentisnotnecessary.How- 3) Cost Comparison: We compare the proposed stochastic
ever, due to the aforementioned challenges and the uncertain model with both the Expected Value Formulation (EVF)
environmentsindistributedquantumcomputing,thedetermin- model and the random model. The EVF model solves the
istic resource allocation approach is not applicable and the deterministic model using the average values of the uncertain
on-demand deployment will be the solution. Therefore, we parameters. The cost of the new quantum computer deploy-
proposetheadaptivedistributedquantumcomputingapproach ment is varied. In the random model, the deployed quantum
based on the two-stage stochastic programming model. The computer in the first stage is randomly selected. Figure 3a
first stage defines the number of deploying the reserved quan- depicts the comparison of total costs of the three models.
tum computers, while the second stage defines the number of We observe that the proposed model achieves the lowest total
installing the on-demand deployment of quantum computers. cost. The minimum total cost cannot be guaranteed by using
The stochastic programming model can be formulated as the the average values of the uncertain parameters used in the
minimization of the total cost including two parts, where the EVF. In addition, the EVF and random models are unable
tum mechanics. According to the increased amount of IoT-
1e4
connected devices, the scalability of utilizing one quantum
Total Cost
8 Dep. QC Cost computer needs to be extended to enhance the overall deploy-
Comp. Cost ment using the properties of distributed quantum computers.
Comm. Cost 3) Future UAV Trajectory Planning: Optimizing UAVs
tsoC 6 Dep. new QC Cost trajectory is challenging when knowledge of ground users,
e.g., locations and channel state information, are unknown.
latoT
4 Quantum-inspired reinforcement learning (QiRL) approach
adopts superposition and amplitude amplification in quantum
mechanics to select probabilistic action and strategy solved
2
by conventional reinforcement learning on classical comput-
ers [7]. The actual quantum computers, including distributed
0
quantum computing, are required to improve convergence
0.0 0.2 0.4 0.6 0.8 1.0 speed and learning effectiveness.
Probability of demand scenarios ( 1) These applications may all be categorized as large-scale
(a) Cost breakdown under different probabilities problems in future networks, which are still challenging be-
cause of the numerous computational resources and processes
1e5
required. A fresh solution to these problems could be further
Proposed model addressed by the distributed quantum computing paradigm.
1.6
EVF model Furthermore, classical and quantum computers will still coex-
1.4 Random model ist to execute computational tasks. Hybrid computing, which
tsoC combines quantum and classical computing, is required to
1.2 significantly reduce energy consumption and costs.